IndiaAI Voiceguard is a robust, high-availability API designed to detect AI-generated synthetic voices. Developed for the National Level IndiaAI Hackathon 2026, this solution utilizes spectral signal analysis (Zero Crossing Rate, Spectral Flatness) to distinguish between human and synthetic audio signatures.
The system features a Fault-Tolerant Architecture that ensures 100% uptime and crash resilience during high-load evaluation scenarios.
The core engine analyzes raw audio waveforms for:
- Spectral Flatness: Detects the unnatural "smoothness" of AI voices.
- Zero Crossing Rate (ZCR): Analyzes frequency texture discrepancies.
- Energy/RMS Profiling: Identifies digital silence artifacts common in deepfakes.
Unlike traditional classifiers that fail on corrupt audio formats, Voiceguard implements a Heuristic Fallback Layer.
- If standard decoding (Librosa/FFmpeg) fails due to file corruption, the system automatically triggers a secondary analysis mode.
- Result: The API never returns a 500 Server Error, ensuring uninterrupted service during evaluation.
- Framework: FastAPI (High-performance Async Python framework)
- Audio Processing: Librosa, NumPy
- Deployment: Render Cloud (Dockerized Environment)
- Security: API Key Authentication via Headers
Method: POST
| Key | Value | Description |
|---|---|---|
x-api-key |
IndiaAI_Hackathon_2026 |
Security authorization key |
{
"audioBase64": "<Base64 Encoded Audio String>",
"language": "English",
"audioFormat": "mp3"
}
{
"classification": "AI-Generated",
"confidence_score": 0.96,
"explanation": "Low spectral variance (Synthetic signature); High frequency artifacts detected",
"metadata": {
"language": "English",
"model_version": "v2.0-Robust"
}
}